Proceedings of the 12th Conference of the European Chapter of the ACL, pages 345–353, Athens, Greece, 30 March – 3 April 2009. c 2009 Association for Computational Linguistics End-to-End Evaluation in Simultaneous Translation Olivier Hamon 1,2 , Christian Fügen 3 , Djamel Mostefa 1 , Victoria Arranz 1 , Muntsin Kolss 3 , Alex Waibel 3,4 and Khalid Choukri 1 1 Evaluations and Language Resources Distribution Agency (ELDA), Paris, France 2 LIPN (UMR 7030) – Université Paris 13 & CNRS, Villetaneuse, France 3 Univerität Karlsruhe (TH), Germany 4 Carnegie Mellon University, Pittsburgh, USA {hamon|mostefa|arranz|choukri}@elda.org, {fuegen|kolss|waibel}@ira.uka.de Abstract This paper presents the end-to-end evalu- ation of an automatic simultaneous trans- lation system, built with state-of-the-art components. It shows whether, and for which situations, such a system might be advantageous when compared to a human interpreter. Using speeches in English translated into Spanish, we present the evaluation procedure and we discuss the results both for the recognition and trans- lation components as well as for the over- all system. Even if the translation process remains the Achilles’ heel of the system, the results show that the system can keep at least half of the information, becoming potentially useful for final users. 1 Introduction Anyone speaking at least two different languages knows that translation and especially simultaneous interpretation are very challenging tasks. A human translator has to cope with the special nature of different languages, comprising phenomena like terminology, compound words, idioms, dialect terms or neologisms, unexplained acronyms or ab- breviations, proper names, as well as stylistic and punctuation differences. Further, translation or in- terpretation are not a word-by-word rendition of what was said or written in a source language. In- stead, the meaning and intention of a given sen- tence have to be reexpressed in a natural and fluent way in another language. Most professional full-time conference inter- preters work for international organizations like the United Nations, the European Union, or the African Union, whereas the world’s largest em- ployer of translators and interpreters is currently the European Commission. In 2006, the European Parliament spent about 300 million Euros, 30% of its budget, on the interpretation and translation of the parliament speeches and EU documents. Gen- erally, about 1.1 billion Euros are spent per year on the translating and interpreting services within the European Union, which is around 1% of the total EU-Budget (Volker Steinbiss, 2006). This paper presents the end-to-end evaluation of an automatic simultaneous translation system, built with state-of-the-art components. It shows whether, and in which cases, such a system might be advantageous compared to human interpreters. 2 Challenges in Human Interpretation According to Al-Khanji et al. (2000), researchers in the field of psychology, linguistics and interpre- tation seem to agree that simultaneous interpre- tation (SI) is a highly demanding cognitive task involving a basic psycholinguistic process. This process requires the interpreter to monitor, store and retrieve the input of the source language in a continuous manner in order to produce the oral rendition of this input in the target language. It is clear that this type of difficult linguistic and cog- nitive operation will force even professional in- terpreters to elaborate lexical or synthetic search strategies. Fatigue and stress have a negative effect on the interpreter, leading to a decrease in simultaneous interpretation quality. In a study by Moser-Mercer et al. (1998), in which professional speakers were asked to work until they could no longer provide acceptable quality, it was shown that (1) during the first 20 minutes the frequency of errors rose steadily, (2) the interpreters, however, seemed to be unaware of this decline in quality, (3) after 60 minutes, all subjects made a total of 32.5 mean- ing errors, and (4) in the category of nonsense the number of errors almost doubled after 30 minutes on the task. Since the audience is only able to evaluate the simultaneously interpreted discourse by its form, 345 the fluency of an interpretation is of utmost im- portance. According to a study by Kopczynski (1994), fluency and style were third on a list of priorities (after content and terminology) of el- ements rated by speakers and attendees as con- tributing to quality. Following the overview in (Yagi, 2000), an interpretation should be as natu- ral and as authentic as possible, which means that artificial pauses in the middle of a sentence, hes- itations, and false-starts should be avoided, and tempo and intensity of the speaker’s voice should be imitated. Another point to mention is the time span be- tween a source language chunk and its target lan- guage chunk, which is often referred to as ear- voice-span. Following the summary in (Yagi, 2000), the ear-voice-span is variable in duration depending on some source and target language variables, like speech delivery rate, information density, redundancy, word order, syntactic charac- teristics, etc. Short delays are usually preferred for several reasons. For example, the audience is irri- tated when the delay is too large and is soon asking whether there is a problem with the interpretation. 3 Automatic Simultaneous Translation Given the explanations above on human interpre- tation, one has to weigh two factors when consid- ering the use of simultaneous translation systems: translation quality and cost. The major disadvantage of an automatic system compared to human interpretation is its translation quality, as we will see in the following sections. Current state-of-the-art systems may reach satis- factory quality for people not understanding the lecturer at all, but are still worse than human inter- pretation. Nevertheless, an automatic system may have considerable advantages. One such advantage is its considerable short- term memory: storing long sequences of words is not a problem for a computer system. Therefore, compensatory strategies are not necessary, regard- less of the speaking rate of the speaker. However, depending on the system’s translation speed, la- tency may increase. While it is possible for hu- mans to compress the length of an utterance with- out changing its meaning (summarization), it is still a challenging task for automatic systems. Human simultaneous interpretation is quite ex- pensive, especially due to the fact that usually two interpreters are necessary. In addition, human in- terpreters require preparation time to become fa- miliar with the topic. Moreover, simultaneous in- terpretation requires a soundproof booth with au- dio equipment, which adds an overall cost that is unacceptable for all but the most elaborate multi- lingual events. On the other hand, a simultaneous translation system also needs time and effort for preparation and adaptation towards the target ap- plication, language and domain. However, once adapted, it can be easily re-used in the same do- main, language, etc. Another advantage is that the transcript of a speech or lecture is produced for free by using an automatic system in the source and target languages. 3.1 The Simultaneous Translation System Figure 1 shows a schematic overview of the si- multaneous translation system developed at Uni- versität Karlsruhe (TH) (Fügen et al., 2006b). The speech of the lecturer is recorded with the help of a close-talk microphone and processed by the speech recognition component (ASR). The par- tial hypotheses produced by the ASR module are collected in the resegmentation component, for merging and re-splitting at appropriate “seman- tic” boundaries. The resegmented hypotheses are then transferred to one or more machine transla- tion components (MT), at least one per language pair. Different output technologies may be used for presenting the translations to the audience. For a detailed description of the components as well as the client-server framework used for connect- ing the components please refer to (Fügen et al., 2006b; Fügen et al., 2006a; Kolss et al., 2006; Fü- gen and Kolss, 2007; Fügen et al., 2001). 3.2 End-to-End Evaluation The evaluation in speech-to-speech translation jeopardises many concepts and implies a lot of subjectivity. Three components are involved and an overall system may grow the difficulty of esti- mating the output quality. However, two criteria are mainly accepted in the community: measuring the information preservation and determining how much of the translation is understandable. Several end-to-end evaluations in speech-to- speech translation have been carried out in the last few years, in projects such as JANUS (Gates et al., 1996), Verbmobil (Nübel, 1997) or TC-STAR (Hamon et al., 2007). Those projects use the main criteria depicted above, and protocols differ in terms of data preparation, rating, procedure, etc. 346 Dictionary Source Hypothesis Translatable Segment Model Source Boundary Resegmen− tationRecognition Speech Translation Model Model Target Language Machine Translation Model Source Acoustic Model Source Language Output Translated Translation Vocabulary Audio Stream Text Output (Subtitles)(Synthesis) Spoken Figure 1: Schematic overview and information flow of the simultaneous translation system. The main components of the system are represented by cornered boxes and the models used for theses components by ellipses. The different output forms are represented by rounded boxes. To our opinion, to evaluate the performance of a complete speech-to-speech translation system, we need to compare the source speech used as input to the translated output speech in the target language. To that aim, we reused a large part of the evalua- tion protocol from the TC-STAR project(Hamon et al., 2007). 4 Evaluation Tasks The evaluation is carried out on the simultaneously translated speech of a single speaker’s talks and lectures in the field of speech processing, given in English, and translated into Spanish. 4.1 Data used Two data sets were selected from the talks and lectures. Each set contained three excerpts, no longer than 6 minutes each and focusing on dif- ferent topics. The former set deals with speech recognition and the latter with the descriptions of European speech research projects, both from the same speaker. This represents around 7,200 En- glish words. The excerpts were manually tran- scribed to produce the reference for the ASR eval- uation. Then, these transcriptions were manually translated into Spanish by two different transla- tors. Two reference translations were thus avail- able for the spoken language translation (SLT) evaluation. Finally, one human interpretation was produced from the excerpts as reference for the end-to-end evaluation. It should be noted that for the translation system, speech synthesis was used to produce the spoken output. 4.2 Evaluation Protocol The system is evaluated as a whole (black box evaluation) and component by component (glass box evaluation): ASR evaluation. The ASR module is evaluated by computing the Word Error Rate (WER) in case insensitive mode. SLT evaluation. For the SLT evaluation, the au- tomatically translated text from the ASR output is compared with two manual reference translations by means of automatic and human metrics. Two automatic metrics are used: BLEU (Papineni et al., 2001) and mWER (Niessen et al., 2000). For the human evaluation, each segment is eval- uated in relation to adequacy and fluency (White and O’Connell, 1994). For the evaluation of ad- equacy, the target segment is compared to a ref- erence segment. For the evaluation of fluency, the quality of the language is evaluated. The two types of evaluation are done independently, but each evaluator did both evaluations (first that of fluency, then that of adequacy) for a certain num- ber of segments. For the evaluation of fluency, evaluators had to answer the question: “Is the text written in good Spanish?”. For the evaluation of adequacy, evaluators had to answer the question: “How much of the meaning expressed in the ref- erence translation is also expressed in the target translation?”. For both evaluations, a five-point scale is pro- posed to the evaluators, where only extreme val- ues are explicitly defined. Three evaluations are carried out per segment, done by three different evaluators, and segments are divided randomly, because evaluators must not recreate a “story” 347 and thus be influenced by the context. The total number of judges was 10, with around 100 seg- ments per judge. Furthermore, the same number of judges was recruited for both categories: ex- perts, from the domain with a knowledge of the technology, and non-experts, without that knowl- edge. End-to-End evaluation. The End-to-End eval- uation consists in comparing the speech in the source language to the output speech in the tar- get language. Two important aspects should be taken into account when assessing the quality of a speech-to-speech system. First, the information preservation is measured by using “comprehension questionnaires”. Ques- tions are created from the source texts (the En- glish excerpts), then questions and answers are translated into Spanish by professional translators. These questions are asked to human judges after they have listened to the output speech in the tar- get language (Spanish). At a second stage, the an- swers are analysed: for each answer a Spanish val- idator gives a score according to a binary scale (the information is either correct or incorrect). This al- lows us to measure the information preservation. Three types of questions are used in order to di- versify the difficulty of the questions and test the system at different levels: simple Factual (70%), yes/no (20%) and list (10%) questions. For in- stance, questions were: What is the larynx respon- sible for?, Have all sites participating in CHIL built a CHIL room?, Which types of knowledge sources are used by the decoder?, respectively. The second important aspect of a speech-to- speech system is the quality of the speech output (hereafter quality evaluation). For assessing the quality of the speech output one question is asked to the judges at the end of each comprehension questionnaire: “Rate the overall quality of this au- dio sample”, and values go from 1 (“1: Very bad, unusable”) to 5 (“It is very useful”). Both auto- matic system and interpreter outputs were evalu- ated with the same methodology. Human judges are real users and native Span- ish speakers, experts and non-experts, but different from those of the SLT evaluation. Twenty judges were involved (12 excerpts, 10 evaluations per ex- cerpt and 6 evaluations per judge) and each judge evaluated both automatic and human excerpts on a 50/50 percent basis. 5 Components Results 5.1 Automatic Speech Recognition The ASR output has been evaluated using the manual transcriptions of the excerpts. The overall Word Error Rate (WER) is 11.9%. Table 1 shows the WER level for each excerpt. Excerpts WER [%] L043-1 14.5 L043-2 14.5 L043-3 9.6 T036-1 11.3 T036-2 11.7 T036-3 9.2 Overall 11.9 Table 1: Evaluation results for ASR. T036 excerpts seem to be easier to recognize au- tomatically than L043 ones, probably due to the more general language of the former. 5.2 Machine Translation 5.2.1 Human Evaluation Each segment within the human evaluation is eval- uated 4 times, each by a different judge. This aims at having a significant number of judgments and measuring the consistency of the human evalua- tions. The consistency is measured by computing the Cohen’s Kappa coefficient (Cohen, 1960). Results show a substantial agreement for flu- ency (kappa of 0.64) and a moderate agreement for adequacy (0.52).The overall results of the hu- man evaluation are presented in Table 2. Regard- ing both experts’ and non-experts’ details, agree- ment is very similar (0.30 and 0.28, respectively). All judges Experts Non experts Fluency 3.13 2.84 3.42 Adequacy 3.26 3.21 3.31 Table 2: Average rating of human evalua- tions [1<5]. Both fluency and adequacy results are over the mean. They are lower for experts than for non- experts. This may be due to the fact that experts are more familiar with the domain and therefore more demanding than non experts. Regarding the detailed evaluation per judge, scores are generally lower for non-experts than for experts. 348 5.2.2 Automatic Evaluation Scores are computed using case-sensitive metrics. Table 3 shows the detailed results per excerpt. Excerpts BLEU [%] mWER [%] L043-1 25.62 58.46 L043-2 22.60 62.47 L043-3 28.73 62.64 T036-1 34.46 55.13 T036-2 29.41 59.91 T036-3 35.17 50.77 Overall 28.94 58.66 Table 3: Automatic Evaluation results for SLT. Scores are rather low, with a mWER of 58.66%, meaning that more than half of the translation is correct. According to the scoring, the T036 ex- cerpts seem to be easier to translate than the L043 ones, the latter being of a more technical nature. 6 End-to-End Results 6.1 Evaluators Agreement In this study, ten judges carried out the evaluation for each excerpt. In order to observe the inter- judges agreement, the global Fleiss’s Kappa co- efficient was computed, which allows to measure the agreement between m judges with r criteria of judgment. This coefficient shows a global agree- ment between all the judges, which goes beyond Cohen’s Kappa coefficient. However, a low co- efficient requires a more detailed analysis, for in- stance, by using Kappa for each pair of judges. Indeed, this allows to see how deviant judges are from the typical judge behaviour. For m judges, n evaluations and r criteria, the global Kappa is defined as follows: κ = 1 − nm 2 −  n i=1  r j=1 X 2 ij nm(m − 1)  r j=1 P j (1 − P j ) where: P j =  n i=1 X ij nm and: X ij is the number of judgments for the i th evaluation and the j th criteria. Regarding quality evaluation (n = 6, m = 10, r = 5), Kappa values are low for both human in- terpreters (κ = 0.07) and the automatic system (κ = 0.01), meaning that judges agree poorly (Landis and Koch, 1977). This is explained by the extreme subjectivity of the evaluation and the small number of evaluated excerpts. Looking at each pair of judges and the Kappa coefficients themselves, there is no real agreement, since most of the Kappa values are around zero. However, some judge pairs show fair agreement, and some others show moderate or substantial agreement. It is observed, though, that some criteria are not fre- quently selected by the judges, which limits the statistical significance of the Kappa coefficient. The limitations are not the same for the com- prehension evaluation (n = 60, m = 10, r = 2), since the criteria are binary (i.e. true or false). Re- garding the evaluated excerpts, Kappa values are 0.28 for the automatic system and 0.30 for the in- terpreter. According to Landis and Koch (1977), those values mean that judges agree fairly. In order to go further, the Kappa coefficients were computed for each pair of judges. Results were slightly better for the interpreter than for the au- tomatic system. Most of them were between 0.20 and 0.40, implying a fair agreement. Some judges agreed moderately. Furthermore, it was also observed that for the 120 available questions, 20 had been answered correctly by all the judges (16 for the interpreter evaluation and 4 for the automatic system one) and 6 had been answered wrongly by all judges (1 for the former and 5 for the latter). That shows a trend where the interpreter comprehension would be easier than that of the automatic system, or at least where the judgements are less questionable. 6.2 Quality Evaluation Table 4 compares the quality evaluation results of the interpreter to those of the automatic system. Samples Interpreter Automatic system L043-1 3.1 1.6 L043-2 2.9 2.3 L043-3 2.4 2.1 T036-1 3.6 3.1 T036-2 2.7 2.5 T036-3 3.5 2.5 Mean 3.03 2.35 Table 4: Quality evaluation results for the inter- preter and the automatic system [1<5]. As can be seen, with a mean score of 3.03 even for the interpreter, the excerpts were difficult to interpret and translate. This is particularly so for 349 L043, which is more technical than T036. The L043-3 excerpt is particularly technical, with for- mulae and algorithm descriptions, and even a com- plex description of the human articulatory system. In fact, L043 provides a typical presentation with an introduction, followed by a deeper description of the topic. This increasing complexity is re- flected on the quality scores of the three excerpts, going from 3.1 to 2.4. T036 is more fluent due to the less technical na- ture of the speech and the more general vocabu- lary used. However, the T036-2 and T036-3 ex- cerpts get a lower quality score, due to the descrip- tion of data collections or institutions, and thus the use of named entities. The interpreter does not seem to be at ease with them and is mispronounc- ing some of them, such as “Grenoble” pronounced like in English instead of in Spanish. The inter- preter seems to be influenced by the speaker, as can also be seen in his use of the neologism “el ce- nario” (“the scenario”) instead of “el escenario”. Likewise, “Karlsruhe” is pronounced three times differently, showing some inconsistency of the in- terpreter. The general trend in quality errors is similar to those of previous evaluations: lengthening words (“seeeeñales”), hesitations, pauses between syl- lables and catching breath (“caracterís ticas”), careless mistakes (“probibilidad” instead of “prob- abilidad”), self-correction of wrong interpreting (“reconocien-/reconocimiento”), etc. An important issue concerns gender and num- ber agreement. Those errors are explained by the presence of morphological gender in Spanish, like in “estos señales” instead of “estas señales” (“these signals”) together with the speaker’s speed of speech. The speaker seems to start by default with a masculine determiner (which has no gen- der in English), adjusting the gender afterward de- pending on the noun following. A quick transla- tion may also be the cause for this kind of errors, like “del señal acustico” (“of the acoustic signal”) with a masculine determiner, a feminine substan- tive and ending in a masculine adjective. Some translation errors are also present, for instance “computerizar” instead of “calcular” (“compute”). The errors made by the interpreter help to un- derstand how difficult oral translation is. This should be taken into account for the evaluation of the automatic system. The automatic system results, like those of the interpreter, are higher for T036 than for L043. However, scores are lower, especially for the L043-1 excerpt. This seems to be due to the type of lexicon used by the speaker for this ex- cerpt, more medical, since the speaker describes the articulatory system. Moreover, his description is sometimes metaphorical and uses a rather col- loquial register. Therefore, while the interpreter finds it easier to deal with these excerpts (known vocabulary among others) and L043-3 seems to be more complicated (domain-specific, technical as- pect), the automatic system finds it more compli- cated with the former and less with the latter. In other words, the interpreter has to “understand” what is said in L043-3, contrary to the automatic system, in order to translate. Scores are higher for the T036 excerpts. In- deed, there is a high lexical repetition, a large number of named entities, and the quality of the excerpt is very training-dependant. However, the system runs into trouble to process foreign names, which are very often not understandable. Differ- ences between T036-1 and the other T036 excerpts are mainly due to the change in topic. While the former deals with a general vocabulary (i.e. de- scription of projects), the other two excerpts de- scribe the data collection, the evaluation metrics, etc., thus increasing the complexity of translation. Generally speaking, quality scores of the au- tomatic system are mainly due to the transla- tion component, and to a lesser extent to the recognition component. Many English words are not translated (“bush”, “keyboards”, “squeaking”, etc.), and word ordering is not always correct. This is the case for the sentence “how we solve it”, translated into “cómo nos resolvers lo” instead of “cómo lo resolvemos”. Funnily enough, the problems of gender (“maravillosos aplicaciones” - masc. vs fem.) and number (“pueden real- mente ser aplicado” - plu. vs sing.) the in- terpreter has, are also found for the automatic system. Moreover, the translation of compound nouns often shows wrong word ordering, in partic- ular when they are long, i.e. up to three words (e.g. “reconocimiento de habla sistemas” for “speech recognition system” instead of “sistemas de re- conocimiento de habla”). Finally, some error combinations result in fully non-understandable sentences, such as: “usted tramo se en emacs es squeaking ruido y dries todos demencial” 350 where the following errors take place: • tramo: this translation of “stretch” results from the choice of a substantive instead of a verb, giving rise to two choices due to the lex- ical ambiguity: “estiramiento” and “tramo”, which is more a linear distance than a stretch in that context; • se: the pronoun “it” becomes the reflexive “se” instead of the personal pronoun “lo”; • emacs: the recognition module transcribed the couple of words “it makes” into “emacs”, not translated by the translation module; • squeaking: the word is not translated by the translation module; • dries: again, two successive errors are made: the word “drives” is transcribed into “dries” by the recognition module, which is then left untranslated. The TTS component also contributes to decreas- ing the output quality. The prosody module finds it hard to make the sentences sound natural. Pauses between words are not very frequent, but they do not sound natural (i.e. like catching breath) and they are not placed at specific points, as it would be done by a human. For instance, the prosody module does not link the noun and its determiner (e.g. “otros aplicaciones”). Finally, a not user- friendly aspect of the TTS component is the rep- etition of the same words always pronounced in the same manner, what is quite disturbing for the listener. 6.3 Comprehension Evaluation Tables 5 and 6 present the results of the compre- hension evaluation, for the interpreter and for the automatic system, respectively. They provide the following information: identifiers of the excerpt: Source data are the same for the interpreter and the automatic system, namely the English speech; subj. E2E: The subjective results of the end-to- end evaluation are done by the same assessors who did the quality evaluation. This shows the percentage of good answers; fair E2E: The objective verification of the an- swers. The audio files are validated to check whether they contain the answers to the ques- tions or not (as the questions were created from the English source). This shows the maximum percentage of answers an evalua- tor managed to find from either the interpreter (speaker audio) or the automatic system out- put (TTS) in Spanish. For instance, informa- tion in English could have been missed by the interpreter because he/she felt that this in- formation was meaningless and could be dis- carded. We consider those results as an ob- jective evaluation. SLT, ASR: Verification of the answers in each component of the end-to-end process. In or- der to determine where the information for the automatic system is lost, files from each component (recognised files for ASR, trans- lated files for SLT, and synthesised files for TTS in the “fair E2E” column) are checked. Excerpts subj. E2E fair E2E L043-1 69 90 L043-2 75 80 L043-3 72 60 T036-1 80 100 T036-2 73 80 T036-3 76 100 Mean 74 85 Table 5: Comprehension evaluation results for the interpreter [%]. Regarding Table 5, the interpreter loses 15% of the information (i.e. 15% of the answers were incorrect or not present in the interpreter’s trans- lation) and judges correctly answered 74% of the questions. Five documents get above 80% of cor- rect results, while judges find almost above 70% of the answers for the six documents. Regarding the automatic system results (Table 6), the information rate found by judges is just above 50% since, by extension, more than half the questions were correctly answered. The lowest excerpt, L043-1, gets a rate of 25%, the highest, T036-1, a rate of 76%, which is in agreement with the observation for the quality evaluation. Infor- mation loss can be found in each component, es- pecially for the SLT module (35% of the informa- tion is lost here). It should be noticed that the TTS module made also errors which prevented judges 351 Excerpts subj. E2E fair E2E SLT ASR L043-1 25 30 30 70 L043-2 62 70 80 70 L043-3 43 40 60 100 T036-1 76 80 90 100 T036-2 61 70 60 80 T036-3 47 60 70 80 Mean 52 58 65 83 Table 6: Comprehension evaluation results for the automatic system [%]. from answering related questions. Moreover, the ASR module loses 17% of the information. Those results are certainly due to the specific vocabulary used in this experiment. So as to objectively compare the interpreter with the automatic system, we selected the questions for which the answers were included in the inter- preter files (i.e. those in the “fair E2E” column of Table 5). The goal was to compare the overall quality of the speech-to-speech translation to in- terpreters’ quality, without the noise factor of the information missing. The assumption is that the interpreter translates the “important information” and skips the useless parts of the original speech. This experiment is to measure the level of this in- formation that is preserved by the automatic sys- tem. So a new subset of results was obtained, on the information kept by the interpreter. The same study was repeated for the three components and the results are shown in Tables 7 and 8. Excerpts subj. E2E fair E2E SLT ASR L043-1 27 33 33 78 L043-2 65 75 88 75 L043-3 37 67 83 100 T036-1 76 80 90 100 T036-2 69 88 75 100 T036-3 47 60 70 80 Mean 53 60 70 80 Table 7: Evaluation results for the automatic sys- tem restricted to the questions for which answers can be found in the interpreter speech [%]. Comparing the automatic system to the inter- preter, the automatic system keeps 40% of the in- formation where the interpreter translates the doc- uments correctly. Those results confirm that ASR loses a lot of information (20%), while SLT loses 10% further, and so does the TTS. Judges are quite close to the objective validation and found most of the answers they could possibly do. Excerpts subj. E2E L043-1 66 L043-2 90 L043-3 88 T036-1 80 T036-2 81 T036-3 76 Mean 80 Table 8: Evaluation results for interpreter, re- stricted to the questions for which answers can be found in the interpreter speech [%]. Subjective results for the restricted evaluation are similar to the previous results, on the full data (80% vs 74% of the information found by the judges). Performance is good for the interpreter: 98% of the information correctly translated by the automatic system is also correctly interpreted by the human. Although we can not compare the performance of the restricted automatic system to that of the restricted interpreter (since data sets of questions are different), it seems that of the inter- preter is better. However, the loss due to subjective evaluation seems to be higher for the interpreter than for the automatic system. 7 Conclusions Regarding the SLT evaluation, the results achieved with the simultaneous translation system are still rather low compared to the results achieved with offline systems for translating European parlia- ment speeches in TC-STAR. However, the offline systems had almost no latency constraints, and parliament speeches are much easier to recognize and translate when compared to the more spon- taneous talks and lectures focused in this paper. This clearly shows the difficulty of the whole task. However, the human end-to-end evaluation of the system in which the system is compared with hu- man interpretation shows that the current transla- tion quality allows for understanding of at least half of the content, and therefore, may be already quite helpful for people not understanding the lan- guage of the lecturer at all. 352 References Rajai Al-Khanji, Said El-Shiyab, and Riyadh Hussein. 2000. On the Use of Compensatory Strategies in Si- multaneous Interpretation. Meta : Journal des tra- ducteurs, 45(3):544–557. Jacob Cohen. 1960. A coefficient of agreement for nominal scales. In Educational and Psychological Measurement, volume 20, pages 37–46. Christian Fügen and Muntsin Kolss. 2007. The influ- ence of utterance chunking on machine translation performance. In Proc. of the European Conference on Speech Communication and Technology (INTER- SPEECH), Antwerp, Belgium, August. ISCA. Christian Fügen, Martin Westphal, Mike Schneider, Tanja Schultz, and Alex Waibel. 2001. LingWear: A Mobile Tourist Information System. In Proc. of the Human Language Technology Conf. (HLT), San Diego, California, March. NIST. Christian Fügen, Shajith Ikbal, Florian Kraft, Kenichi Kumatani, Kornel Laskowski, John W. McDonough, Mari Ostendorf, Sebastian Stüker, and Matthias Wölfel. 2006a. The isl rt-06s speech-to-text system. In Steve Renals, Samy Bengio, and Jonathan Fiskus, editors, Machine Learning for Multimodal Interac- tion: Third International Workshop, MLMI 2006, Bethesda, MD, USA, volume 4299 of Lecture Notes in Computer Science, pages 407–418. Springer Ver- lag Berlin/ Heidelberg. Christian Fügen, Muntsin Kolss, Matthias Paulik, and Alex Waibel. 2006b. Open Domain Speech Trans- lation: From Seminars and Speeches to Lectures. In TC-Star Speech to Speech Translation Workshop, Barcelona, Spain, June. Donna Gates, Alon Lavie, Lori Levin, Alex. Waibel, Marsal Gavalda, Laura Mayfield, and Monika Wosz- cyna. 1996. End-to-end evaluation in janus: A speech-to-speech translation system. In Proceed- ings of the 6th ECAI, Budapest. Olivier Hamon, Djamel Mostefa, and Khalid Choukri. 2007. End-to-end evaluation of a speech-to-speech translation system in tc-star. In Proceedings of the MT Summit XI, Copenhagen, Denmark, September. Muntsin Kolss, Bing Zhao, Stephan Vogel, Ashish Venugopal, and Ying Zhang. 2006. The ISL Statis- tical Machine Translation System for the TC-STAR Spring 2006 Evaluations. In TC-Star Workshop on Speech-to-Speech Translation, Barcelona, Spain, December. Andrzej Kopczynski, 1994. Bridging the Gap: Empiri- cal Research in Simultaneous Interpretation, chapter Quality in Conference Interpreting: Some Pragmatic Problems, pages 87–100. John Benjamins, Amster- dam/ Philadelphia. J. Richard Landis and Gary G. Koch. 1977. The mea- surement of observer agreement for categorical data. In Biometrics, Vol. 33, No. 1 (Mar., 1977), pp. 159- 174. Barbara Moser-Mercer, Alexander Kunzli, and Ma- rina Korac. 1998. Prolonged turns in interpreting: Effects on quality, physiological and psychological stress (pilot study). Interpreting: International jour- nal of research and practice in interpreting, 3(1):47– 64. Sonja Niessen, Franz Josef Och, Gregor Leusch, and Hermann Ney. 2000. An evaluation tool for ma- chine translation: Fast evaluation for mt research. In Proceedings of the 2nd International Conference on Language Resources and Evaluation, Athens, Greece. Rita Nübel. 1997. End-to-end Evaluation in Verb- mobil I. In Proceedings of the MT Summit VI, San Diego. Kishore Papineni, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2001. Bleu: a method for automatic evaluation of machine translation. Technical Report RC22176 (W0109-022), Research Report, Com- puter Science IBM Research Division, T.J.Watson Research Center. Accipio Consulting Volker Steinbiss. 2006. Sprachtechnologien für Europa. www.tc-star. org/pubblicazioni/D17_HLT_DE.pdf. John S. White and Theresa A. O’Connell. 1994. Evaluation in the arpa machine translation program: 1993 methodology. In HLT ’94: Proceedings of the workshop on Human Language Technology, pages 135–140, Morristown, NJ, USA. Association for Computational Linguistics. Sane M. Yagi. 2000. Studying Style in Simultane- ous Interpretation. Meta : Journal des traducteurs, 45(3):520–547. 353 . this kind of errors, like “del señal acustico” (“of the acoustic signal”) with a masculine determiner, a feminine substan- tive and ending in a masculine. that simultaneous interpre- tation (SI) is a highly demanding cognitive task involving a basic psycholinguistic process. This process requires the interpreter